The real cost of self-hosting Sentry
Self-hosted Sentry is genuinely free software, and for some teams it's the right call. But 'free' is a license price, not a total cost. Having run a production self-hosted Sentry ourselves, here is what the bill actually looks like — in hardware, in upgrades, and in the 3 a.m. pages you inherit.
What you're actually operating
Sentry stopped being 'a Django app with Postgres' many years ago. A current self-hosted install runs on the order of fifty containers: the web app and workers, Kafka and Zookeeper, ClickHouse for event analytics, Redis, Postgres, Snuba query services, Relay for ingestion, symbolicator, vroom for profiling, and a mesh of consumers wiring them together.
The honest TCO line items
| Line item | Typical cost | Notes |
|---|---|---|
| Server(s) | $80–$400/mo | A 16 GB box is the entry point; teams with real volume run several. |
| Storage growth | grows monthly | ClickHouse + Kafka retention needs active pruning or the disk fills. |
| Upgrades | 2–6 hrs/month | Sentry ships fast; self-hosted upgrades occasionally involve breaking migration steps. |
| Monitoring the monitor | your pager | When Kafka lags or a consumer dies, error reporting silently stops — and nothing tells you. |
| Engineer time | the big one | A day a month of a senior engineer often costs more than every SaaS plan combined. |
The failure mode nobody prices in
The worst property of self-hosted monitoring is that it fails silently in the same incidents it exists to catch. A full disk, a stuck Kafka consumer or an OOM-killed worker doesn't page you — it just stops ingesting. You find out days later, when someone asks why the dashboard has been quiet. You end up building a second monitoring system to watch the first one.
When self-hosting is the right call
- Hard data-residency requirements — the data legally cannot leave your infrastructure.
- Existing platform team — you already run Kafka and ClickHouse for other workloads, so the marginal operational cost is genuinely low.
- Massive volume with cheap hardware — at extreme event volumes, owned hardware can beat any per-event SaaS price.
The middle path: keep the SDKs, drop the ops
The trap in this decision is thinking it's self-hosted Sentry or SaaS Sentry. Because the Sentry SDK protocol is open, a third option exists: keep every SDK exactly as it is and point the DSN at a hosted, Sentry-compatible backend. That's what jentry is — hosted error tracking, APM, replays and release health that speaks the Sentry protocol, at a flat price, with no Kafka to babysit.
# the entire migration
- dsn: "https://key@your-sentry-box.internal/1"
+ dsn: "https://key@jentry.app/1"We did exactly this migration ourselves — jentry's own production apps report to jentry, and the self-hosted Sentry install they came from is powered off. The SDKs never changed.
Frequently asked questions
Is self-hosted Sentry really free?
The software is free (FSL-licensed, source available). The cost is the infrastructure it runs on and the engineering time to keep it healthy and upgraded — commonly hundreds of dollars a month in practice.
How much hardware does self-hosted Sentry need?
The documented minimum is 4 CPU cores and 16 GB RAM for the ~50-container stack. Production installs with real traffic and retention typically need more, plus disk that grows continuously.
What breaks most often?
In our experience: disks filling from ClickHouse/Kafka retention, consumers lagging after restarts, and upgrades that require manual migration steps. Each failure stops ingestion silently.
Can I move off self-hosted Sentry without touching my apps?
Yes. Sentry SDKs send to whatever DSN you configure. Point the DSN at a compatible backend like jentry and errors, transactions and releases keep flowing — no code changes.
Try jentry free
Hosted error tracking & performance monitoring. Works with your Sentry SDKs — send your first event in minutes.